Digital rock technology has gradually developed into an essential approach to studying petrophysical properties. As the initial part of the overall digital rock workflow, the reliability and effectiveness of digital rock modeling based on instrument scanning images would affect the accuracy of almost all subsequent numerical simulation and petrophysical analyses. However, the single imaging instrument’s scanning field and resolution are contradictory, i.e., a larger field of view translates into a lower resolution, making it difficult for the quality of the digital rock model to meet research requirements. For this issue, we proposed a deep-learning-based method to integrate the scanning results of diverse imaging instruments and construct the digital rock model with high-resolution capability and a wide field of view. Firstly, we utilized diverse imaging methods to scan the core sample to obtain the dataset for training, validation, and testing of the deep neural network. Then, we used deep-learning techniques to establish the mapping relation between the voxel of low-resolution X-ray computed tomography images for the plunger core sample (plunger CT images) and the various component ratios of high-resolution digital rock. Finally, the trained deep neural network is applied to the whole plunger CT image processing, and the multi-scale and multi- component integration (MSMCI) digital rock model is constructed. In the test dataset, the average absolute error (MAE) between the prediction minerals&pore ratios by the proposed method and the high-resolution labels is 0.12v/v, which is lower than 0.17v/v based on the existing deep-learning super-resolution processing approach, and 0.24v/v based on the multi-scale integration in a statistical approach. In practical applications, the porosity and mineral ratios extracted from the MSMCI plunger digital rock model are consistent with those obtained from laboratory petrophysical experiments. The ability of the MSMC plunger digital rock to characterize multi-scale components inside the rock is significantly better than that of the digital rock segmented from single imaging instrument scanning images.
Wang et al. (Sat,) studied this question.